Using Crowdsourcing to Identify a Proxy of Socio-Economic status

TL;DR

This paper explores using social media community language patterns, derived from discussion forums, as proxies for socio-economic status to aid health and urban service planning.

Contribution

It introduces a method to group users by geographic background through community vocabulary analysis, providing a new approach to estimate socio-economic status from online discussions.

Findings

Distinct vocabulary patterns correlate with different communities.

Language differences can predict socio-economic status.

Framework supports better targeting for health and urban services.

Abstract

Social Media provides researchers with an unprecedented opportunity to gain insight into various facets of human life. Health practitioners put a great emphasis on pinpointing socioeconomic status (SES) of individuals as they can use to it to predict certain diseases. Crowdsourcing is a term coined that entails gathering intelligence from a user community online. In order to group the users online into communities, researchers have made use of hashtags that will cull the interest of a community of users. In this paper, we propose a mechanism to group a certain group of users based on their geographic background and build a corpus for such users. Specifically, we have looked at discussion forums for some vehi-cles where the site has established communities for different areas to air their grievances or sing the praises of the vehicle. From such a discussion, it was pos-sible to glean the vocabulary that these group of users adheres to. We compared the corpus of different communities and noted the difference in the choice of language. This provided us with the groundwork for predicting the socio-eco-nomic status of such communities that can be particularly helpful to health prac-titioners and in turn used in smart cities to provide better services to the commu-nity members. More work is underway to take words and emojis out of vo-cablary(OOV) and assessing the average score as special cases.